AI to Consume 12 Percent of Electricity, but There Are Caveats

A new report paints a dire picture of the future in which electricity demand is surging and the transition to clean energy is still decades away.

The global risk management provider DNV forecasts that global emissions will reach net zero only after 2090, and anticipates a temperature increase of roughly 2.2 degrees Celsius above preindustrial levels by 2100, although they caution it could be higher than that. Furthermore, AI data centers are expected to consume about 12 percent of all electricity in North America as soon as 2040. 

“A casual observer might conclude that the energy transition is stalled or in reverse. That is most definitely not the case,” the report states. “Some aspects of the transition are supercharged and progressing rapidly, while other aspects of the transition have hit turbulence and are delayed.”

There are, however, some caveats and reasons for optimism. DNV’s report notes, for example, that as soon as about 2060, carbon dioxide emissions are expected to fall by about 63 percent, with fossil fuels all but exiting the global energy mix. It also states that upheaval in U.S. policy will likely slow the clean energy transition, but not entirely derail it, largely because of China’s leadership in technological development and renewables buildout. 

As far as AI, the report anticipates data centers will account for an outsized chunk of electricity in North America—consuming some 16 percent overall, with that aforementioned 12 percent coming strictly from AI. Globally, however, data centers are expected to surge to consume some 5 percent of electricity by 2040 with 3 percent attributable to AI, specifically. The report, however, anticipates that initial exponential growth in power demand from AI will become linear in time, even as the “cognitive services” it provides grow exponentially. The energy efficiency of “leading [machine learning] hardware” has improved about 40 percent year-over-year, according to the report.

During NYC Climate Week in late September, Nvidia’s head of sustainability Josh Parker joined panels to discuss AI and sustainability. On one, he argued it’s worth it to bring new energy online to fuel AI, if artificial intelligence is applied to accelerate innovation in sustainability, emissions reduction, and clean energy.

“AI really can be—and will be if we use it properly—a fantastic solution to some of the biggest challenges that we’ve had in sustainability,” he said on a separate panel on the same subject. “AI is not only providing more performance per watt of energy, but it’s also more performance per liter of water. It’s more performance per ton of steel, more performance per chip, and across every metric you can think of.”